End-to-End AI Policies for Timing Cross-Asset Futures

Summary

The document examines whether models that map market conditions directly to portfolio weights can outperform conventional cross-asset futures strategies. It describes training these policies on sixteen liquid CME futures with a differentiable Sharpe-ratio objective, then comparing them with equal weighting, risk parity, and time-series momentum. This setup skips the usual separate steps of forecasting returns and optimizing allocations.

The learned policies rank ahead of rule-based benchmarks for the pooled portfolio and some asset groups, though the advantage is not consistent across all groups. Out of sample, an LSTM and a transformer have similar gross performance, but trading costs change the comparison: the transformer trades less and performs better relative to the LSTM, matching or surpassing equal weighting under moderate costs. The excerpt gives no detailed performance statistics, test period, or broader robustness checks, so it does not establish that these results generalize beyond the studied futures and evaluation conditions.

Key ideas

• The method learns portfolio weights directly from market states rather than forecasting returns before optimization.
• Training uses a differentiable Sharpe-ratio objective on a set of liquid cross-asset futures.
• Benchmarks include equal weighting, risk parity, and time-series momentum.
• Learned policies rank well for the combined portfolio and selected asset groups, but results vary by group.
• Transaction costs favor the less active transformer policy over the LSTM in the reported comparison.

Tags

# End-to-End Parametric Portfolio Policies for Cross-Asset Futures Timing: When Do AI Models Beat Simple Rules?


# End-to-End Parametric Portfolio Policies for Cross-Asset Futures Timing: When Do AI Models Beat Simple Rules?









Timing-based tilts across asset classes can drive much of the risk and return of a diversified cross-asset portfolio. The standard approach forecasts returns and then optimizes weights. We instead study an end-to-end AI-based policy that maps market states directly to portfolio weights, and we then ask when this one-step modeling approach outperforms simple rules-based strategies. We train these policies on the sixteen most liquid CME futures, where an edge is unlikely to be due to illiquidity, using a differentiable Sharpe ratio loss function, and we benchmark them against equal weighting, risk parity, and time-series momentum. The learned policies rank above the rules on the pooled cross-asset portfolio and in several sub-asset classes, but not uniformly. In gross terms, an LSTM and a transformer-based architecture perform comparably out-of-sample, but diverge when we consider transaction costs. The transformer generates the stronger learned policy, trades far less than the LSTM, and matches or exceeds equal weighting through moderate cost.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.